In a data center that includes one or more top of rack (TOR) switches connected to end-point server(s), link based failure and load balancing is typically performed by using mechanisms such as simple network management protocol (SNMP) traps for link failure notification, equal-cost multi-path routing (ECMP) for finding the next best path, and border gateway protocol (BGP) for routing (e.g., finding and disseminating network paths to any desired destination). By contrast, a rack scale-based pod switch environment (e.g., a set of compute, network, storage, and/or other resources) may eliminate TOR switches, and instead, each end point network interface controller (NIC) may be connected to two or more switches at the end of a row. In such a configuration, load balancing using SNMP, ECMP, and BGP, and the like are not as effective. In the pod switch environment, load balancing after link failure is especially important in order to maintain a target quality of service.